MADISON, Wisc. Survival of the fittest principles have been employed by researchers at the University of Wisconsin's Engine Research Center (ERC) here to evolve the world's most efficient truck engine one that pollutes less while consuming less fuel.
"Usually you have to optimize for either fuel efficiency or for low pollution levels, but the genetic algorithm was able to improve on both," said researcher Peter Senecal at the ERC. Diesel truck engines must meet strict new emission standards by 2002 that require improved performance as well as lower emissions and fuel consumption.
Senecal developed his genetic algorithm at the ERC with funds supplied by Caterpillar Inc. (Peoria, Ill.). The new design cut nitric-oxide emissions by three times and soot emissions by 50 percent while simultaneously reducing fuel consumption by 15 percent.
The genetic algorithm used by Senecal sifted through billions of combinations of engine parameters that engineers have never had the time to try. After examining and comparing them, the algorithm selected 250 combinations of parameters that were better than conventional designs but were never used to build an engine.
In the final design, only six engine parameters were used: fuel-injection timing, fuel-injection pressure, intake boost pressure, two split-injection velocity curve parameters (percentage in first squirt and delay time between squirts) and the level of exhaust recirculation.
The simulation, run on Sun Microsystems workstations at the ERC, compared all the combinations for all six parameters using natural selection. First, the best-known values for the six engine parameters were put into the model. The performance was measured with a function that minimized both emissions and fuel. Then a population of random parameter values was compared with the known best "baseline" set of parameters. Those comparing most favorably were allowed to "mate" by swapping some of the individual parameter settings.
"We used crossover to cut the parameter sets just like they were strands of DNA. I actually have a screen where I can clip two six-unit strands and attach the head of one onto the tail of the other. Then I used local mutations to perturb the individual parameter settings enough to keep out of local minimums," said Senecal.
This "offspring" population of parameter settings was then tested against the same performance measures as before, and the process continued until the best-known baseline design was left in the dust by the newly evolved engine designs.
"We knew our design looked good on paper, but we tested it on a real engine at the ERC to make sure, and we found excellent confirmation of our simulation results," said Senecal.
Senecal's research has been partially inspired by Caterpillar, which plans to build prototype diesel engines based on the newly discovered design principles, said Senecal. "Engine design is incredibly complex, and when I first discovered genetic algorithms I didn't know if they were robust enough to meet the challenge, but now I think it would have taken decades of conventional research to engineer the improvements that genetic algorithms discovered in just a few weeks," he said.
Senecal plans to spend the next year using genetic algorithms to test different engine geometries such as size, piston shape and other geometric parameters. He hopes to discover new shapes and sizes of engine geometries that have been unexplored by conventional designers but are easy to build.
"We can measure the performance of engine designs evolved by the genetic algorithm using computational fluid dynamics, but the big hurdle will be inventing a technique to automatically generate those engine designs from a set of geometric engine parameters," Senecal said.